This project is read-only.

CUDAfy V1.27 Installer

Rating:        Based on 6 ratings
Reviewed:  6 reviews
Downloads: 1769
Released: Jul 25, 2014
Updated: Jul 25, 2014 by NickKopp
Dev status: Stable Help Icon

Recommended Download

Application CUDAfy V1.27 Installer
application, 2644K, uploaded Jul 25, 2014 - 1769 downloads

Release Notes

Hybrid DSP Systems Signed Windows Installer

Fix: CUDA and OpenCL tests should handle devices with max block threads < 1024
Fix: Single floating point Execute called instead of double (emulator only)
Add: CUDA 6.0 support.
Add: Compute capability 5.0 support.

Reviews for this release

Valuable toolkit which makes CUDA and OpenCL integration into C# development effortless. I've been using Cudafy for a couple of years for my projects, so I can say with confidence that the support that the developers give is extremely good. The forums is also a place where both the developers and the community are very active and keen on aiding people with any issues. Keep up the good work guys!
by ethanbayne on Oct 5, 2014 at 12:31 PM
its made an invaluable addition to my toolset. keep up the fantastic work. i am working with vs2013 and 6.5 (albeit without math since it asks for 6.0 libs).
by terraslate on Sep 22, 2014 at 2:48 PM
CUDAfy is just great, I've been using it in my open-source Chemical Process Simulator, DWSIM. As far as I know, it is the only process simulator with GPU computing capabilites thanks to this library. Keep up the good work!
by DanWBR on Nov 24, 2014 at 3:01 PM
Good tool, unfortunate it doesnt work in vs2013 or with cuda 6.5.
by Phyyl on Sep 7, 2014 at 12:33 AM
What issues are you getting with these? Can you please follow up on discussion forum? Other than the Math libraries it should be fine with CUDA 6.5.
Reply by NickKopp (N/A) on Sep 7, 2014 at 4:46 PM
CUDAfy is nice in concept, but it lack maturity in a few ways: It's very laggy to the CUDA/NVIDIA release cycle, v1.27 which was just release for CUDA 6.0 doesn't work on CUDA 6.5 for no good reason except for (IMO) poor code and hardcoding of paths The project doesn't build out of the box. Not only that, the documentation clearly states it cannot be built by contributors because it uses a 3rd party tool (and it isn't even disclosed which it is!) to package into a single .DLL. I hate to sound complain, but the CUDAfy project could definitely use the community, but the decisions taken and the current quality of the code+build leave it in a place where fixing bugs by the community is needlessly hard (I would love to fix/add/work on CUDA 6.5 support, but the thing cannot be built without guru meditation) I'd love for the CUDAfy people to get their act together, move to github, integrate with some ci tool (like travis + appveyor), do a completely open-source/free build where contributors are actually capable of helping the project
by damageboy on Sep 8, 2014 at 7:35 AM
There have been many contributors over the years from Australia, Korea, Portugal, Canada, Ireland, USA, etc. Once we've had contact they are given access to our internal SVN. The 3rd party tool is .NET Reactor which is a paid for tool used to merge the DLLs. You don't need it for own use since you can link to the discrete DLLs. This is a complex project relying on changing 3rd party software (CUDA, Visual Studio and .NET) & requires learning basics of GPU programming. You don't appear willing to even use the forum to discuss issues.
Reply by NickKopp (N/A) on Sep 8, 2014 at 9:10 AM
For a short term, it is surely bad. It works if your C# code is just like its examples. you may expect few basic features supported since it says it supports C# but you cannot use CLR, nested-classes, nonstatic members, methods, ... and so on. No benefits from C#. For now if someone asks me to use cudafy, I'd rather write pure CUDA code since cudafy's doing mostly about copying your static kernel function's body into CUDA structured source code string(though not raw copy). What I wanted to do is giving GPU computing power to my machine learning code written in C#. It made me decide to rewrite my code in C++AMP. Furthermore it's under LGPL. yup, it deserves 1 star.
by laie0 on Aug 19, 2014 at 7:46 AM
We made the tool for internal use and decided to release to public which cost us a lot of work but it's giving back and has been of use to many people. CUDAfy is two things: a .NET wrapper for the host CUDA functions and a means of writing kernels in .NET. This objective is met to a good extent. You apparently have completely different expectations. This does not necessitate a one star review, it simply requires you to find another tool. LGPL because some (in finance) are making a lot of money out of CUDAfy - to avoid paying all we ask is that you provide any code changes you make. Any proceeds made (not much unfortunately) go to a charity helping handicapped kids in Dharamsala, India.
Reply by NickKopp (N/A) on Aug 19, 2014 at 9:08 AM